Detectability of controlled leaks in an Andean water transmission main: A field evaluation of changepoint methods on flow and pressure signals

Leak detection on pumped transmission mains that cross steep terrain relies on hydraulic signals whose response to a small leak may be comparable to sensor noise. We evaluated five unsupervised methods (PELT with Bayesian and Akaike penalties, Bayesian online changepoint detection [BOCD], an OLS-CUSUM structural-break procedure and a moving-average ±3σ chart) on flow and pressure series recorded every 30 s during seven controlled leak trials (valve setting 3-5% of nominal flow) in two field campaigns on an Andean transmission main in Loja, Ecuador. Detections were compared with a chance-level reference obtained by placing the same number of changepoints at random. Relative to the preceding leak-free observations, flow decreased in all seven trials (mean -0.50%; sign test p = 0.016; placebo test p = 0.003), whereas pressure, recorded at 0.1-m resolution, showed no change (placebo p = 0.99). No single event exceeded its per-event detection threshold; the flow change was resolvable only as a seven-event mean (signal-to-noise ratio -3.3). None of the five detectors located leak onsets more often than chance (pooled p > 0.40 for every method-signal pair). The PELT variants flagged six of seven events on pressure, but at 560-960 false alarms per day and 13-33% precision. Although BOCD on pressure reached the highest average precision (0.54), its excess over the chance reference was not significant after Holm correction (adjusted p = 0.14). A CUSUM tuned to the observed flow change, the optimal sequential detector for a persistent shift, would need 12-29 min to detect it at one false-alarm per day, 8-20 times the trial duration. Chance-level performance is therefore the expected outcome for 90-120-s leaks at 30-s sampling, whereas a persistent leak producing the same flow change would be detectable within tens of minutes.

Authors

Institutions

Publication Details

Journal
PLoS ONE
Published
2026-11-16
DOI
https://doi.org/10.1371/journal.pone.0362980
Primary Topic
Water Systems and Optimization
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Detectability of controlled leaks in an Andean water transmission main: A field evaluation of changepoint methods on flow and pressure signals

Holger Manuel Benavides-Muñoz, Peter Cheung
PLoS ONE
Water Systems and Optimization
article

Detectability of controlled leaks in an Andean water transmission main: A field evaluation of changepoint methods on flow and pressure signals

Holger Manuel Benavides-Muñoz, Peter Cheung
article en

Abstract

Leak detection on pumped transmission mains that cross steep terrain relies on hydraulic signals whose response to a small leak may be comparable to sensor noise. We evaluated five unsupervised methods (PELT with Bayesian and Akaike penalties, Bayesian online changepoint detection [BOCD], an OLS-CUSUM structural-break procedure and a moving-average ±3σ chart) on flow and pressure series recorded every 30 s during seven controlled leak trials (valve setting 3-5% of nominal flow) in two field campaigns on an Andean transmission main in Loja, Ecuador. Detections were compared with a chance-level reference obtained by placing the same number of changepoints at random. Relative to the preceding leak-free observations, flow decreased in all seven trials (mean -0.50%; sign test p = 0.016; placebo test p = 0.003), whereas pressure, recorded at 0.1-m resolution, showed no change (placebo p = 0.99). No single event exceeded its per-event detection threshold; the flow change was resolvable only as a seven-event mean (signal-to-noise ratio -3.3). None of the five detectors located leak onsets more often than chance (pooled p > 0.40 for every method-signal pair). The PELT variants flagged six of seven events on pressure, but at 560-960 false alarms per day and 13-33% precision. Although BOCD on pressure reached the highest average precision (0.54), its excess over the chance reference was not significant after Holm correction (adjusted p = 0.14). A CUSUM tuned to the observed flow change, the optimal sequential detector for a persistent shift, would need 12-29 min to detect it at one false-alarm per day, 8-20 times the trial duration. Chance-level performance is therefore the expected outcome for 90-120-s leaks at 30-s sampling, whereas a persistent leak producing the same flow change would be detectable within tens of minutes.

PLoS ONEVol. 21(11)
Universidad Técnica Particular de Loja (EC), Universidade Federal de Mato Grosso do Sul (BR)
Universidad Técnica Particular de Loja, Universidade Federal de Mato Grosso do Sul
Water Systems and Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon